Federated Learning for Privacy-Preserving Medical Data Analytics in Big Data
Jhansi Bharathi Madavarapu, Ankita Nainwal, Ammar Hameed Shnain, Anurag Shrivastava, Kanchan Yadav, A L N Rao · 2024
This examination investigates the application of Federated Learning (FedAvg, FedSGD, FedProx, HEFL) for security safeguarding medical data analytics about big data. Utilizing a different dataset containing electronic well-being records, medical imaging, and wearable gadget data, our review means to work out some kind of harmony between tackling the experiences implanted in broad medical care data and defending patient protection. Through thorough trials, it looked at the exhibition of federated learning calculations, taking into account measurements, for example, intermingling rate, exactness, and correspondence above. Results uncovered that FedSGD displayed the quickest combination (50 rounds) while keeping a high exactness of$89\%$. FedAvg, described by straightforwardness, exhibited combination in 100 rounds with an exactness of$87\%$. FedProx accomplished a harmony between combination speed (75 adjusts) and model precision ($88\%$). The application of Homomorphic Encryption for Federated Learning (HEFL) showed promising outcomes, meeting in 50 rounds with a precision of$85\%$, despite the higher correspondence above because of encryption. The near examination with related work featured the predominance of federated learning in security safeguarding and productive medical data analytics. This examination contributes significant experiences to the talk on secure and cooperative medical care analytics in the period of big data.